From mboxrd@z Thu Jan  1 00:00:00 1970
From: Mark Mitchell <mark@codesourcery.com>
To: rth@cygnus.com
Cc: binutils@sourceware.cygnus.com
Subject: Re: .dynsym pain
Date: Wed, 14 Jul 1999 01:03:00 -0000
Message-id: <19990714010018W.mitchell@codesourcery.com>
References: <19990713191248.A4175@cygnus.com> <19990713224423R.mitchell@codesourcery.com> <19990714000728.D7058@cygnus.com>
X-SW-Source: 1999-q3/msg00197.html

>>>>> "Richard" == Richard Henderson <rth@cygnus.com> writes:

    Richard> On Tue, Jul 13, 1999 at 10:44:23PM -0700, Mark Mitchell
    Richard> wrote:
    >> Is MIPS really broken, or do you just think it is?

    Richard> I know it is.

OK.

    Richard> I'm doubly convinced another sort is needed.

OK.  In that case, your patch looks reasonable to me, I guess.

But, I'm nervous about the need for *two* sorts.  The reason for the
first sort is to calculate the size of the GOT:

 	  /* There has to be a global GOT entry for every symbol with
 	     a dynamic symbol table index of DT_MIPS_GOTSYM or
 	     higher.  Therefore, it make sense to put those symbols
 	     that need GOT entries at the end of the symbol table.  We
 	     do that here.  */
 	  if (!mips_elf_sort_hash_table (info))
 	    return false;

 	  i = elf_hash_table (info)->dynsymcount - g->global_gotsym->dynindx;
	  s->_raw_size += i * MIPS_ELF_GOT_SIZE (dynobj);

So, if we resort, and the GOT gets bigger, that's very bad.  I assume
that this can't happen.  Perhaps you could add an assertion to check
that this situation does not occur?  Just checking `elf_hash_table
(info)->dynsymcount - g->global_gotsym->dynindx' before and after the
second sort should do it.

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com